Absolutely the same issue here. 40300 rows are missing from the provided test data. Hope someone is gonna answer soon.

What I do:

break the test dataset in two parts: with nulls and without nulls

testOK   = test[~test['customer_id'].isnull()]

testNull = test[test['customer_id'].isnull()]

print (test.shape)

print (testOK.shape)

print (testNull.shape)

After that I got the customer_id, loc_num and vendor for the CID X LOC_NUM X VENDOR column: (only for testNull)

# fill null columns in test after Merge - for customer_id, location_number_obj and id_obj

for i, row in testNull.iterrows():

if pd.isna(row['customer_id']):

text = row['CID X LOC_NUM X VENDOR']

x1 = text.find(' X ')

x2 = text.find(' X ', x1+2)

text1 = text[0:x1]

text2 = text[x1+2: x2]

text3 = text[x2+2:]

test.at[i,'customer_id']         = text1

test.at[i,'location_number_obj'] = text2

test.at[i,'id_obj']              = text3

Them recreate the test dataset:

frames  = [testOK, testNull]

test    = pd.concat(frames)

After that I apply my null column strategy for that customer, Loc_num or vendor